Uncapped with Jack Altman
· Jack Altman, Tobi Lutke
[00:00:00] S0: You can spend it on big, important changes. [00:00:00] S1: And change management, that’s one of the best ways. I can speed up something that would take years of small culture change or internal training with sometimes a memo. I take that really seriously. It’s not the easiest thing. It leads to more work or sometimes unpleasantness. But I find that is what I owe the company. It’s my best way to help the mission. So when something like the AI thing becomes true, we are saying, “Hey, we have two people. They were both equally good programmers 15 minutes ago, but one of them has completely jumped onto the AI train.” Back when I wrote this, it was really hard to actually get real value out of AI. It was more that the premise of AI was important. What do we do in performance reviews? The moment we said it… It’s an impact rating. It’s called net impact reviews at Shopify. What’s your net impact on the company and the mission? It’s just very demonstrably true that one of the people was of more impact. The moment that is said, it feels incredibly unkind not to tell people. So, “Hey, let’s write it out and send it to everyone.” [00:00:00] S0: It’s really unfair if you’re like, “This person has the exoskeleton on, and I’m not going to point out that the exoskeleton’s making him 10 times faster.”
· Jack Altman, Tobi Lutke
[00:00:00] S0: It’s really unfair if you’re like, “This person has the exoskeleton on, and I’m not going to point out that the exoskeleton’s making him 10 times faster.” [00:00:00] S1: If we act on this thing, we should tell people. And so we did. I included a bunch of other things that are true. I’ve invested lots in making sure that everyone has everything they need. We have an unlimited token policy that I’m sure your brother is thrilled with. We want people to tinker. We want people to play with this. We want people to use it a lot. It’s super good. It skyrocketed the second this message came along. What Shopify did with this, the speed of diffusion of this tool was remarkable. I’d like to think Shopify is predisposed for this. We have “thriving on change” as one of our core values, and we really mean it. [00:00:00] S0: It’s good that you talked about net impact. One of the blunter instruments that’s getting used a lot right now is just tracking token consumption. People are like, “I want to see token consumption go up 20% a month.” There are founders who say that. It’s not an obviously terrible idea, in the same way that judging people by lines of code… We could debate, is there anything to it? Probably something, but…
· Jack Altman, Tobi Lutke
[00:00:00] S0: It’s good that you talked about net impact. One of the blunter instruments that’s getting used a lot right now is just tracking token consumption. People are like, “I want to see token consumption go up 20% a month.” There are founders who say that. It’s not an obviously terrible idea, in the same way that judging people by lines of code… We could debate, is there anything to it? Probably something, but… [00:00:00] S1: I think it’s fine. At some point we had a leaderboard and so on of token consumption. Of course, that leads immediately to bad effects, so we don’t have that. On Vault, which is our internal system—it’s probably not a term that anyone’s ever heard, but it’s our internal wiki and everything—on your profile, it shows what your token usage is and which percentile you are in your department and group and so on. Just because that’s interesting, because we are tracking it. We have to, because at some point we have to allocate finances to OpEx and whatnot. Therefore, we show it to people so that they know. [00:00:00] S0: This is probably different numbers because you’re at such a scale, but there are private companies whose token spend as a percentage of revenue is going into pretty wild places right now. I think that’s fine if you’re an earlier stage company and you’re just trying to win the market at all costs. But I am curious how this might play out. I’m sure even you’re seeing token spend at numbers that are probably not a huge deal relative to your revenue, but— [00:00:00] S1: It’s a huge deal compared to revenue. It’s unbelievable.
· Tobi Lutke, Jack Altman
[00:00:00] S1: It’s a huge deal compared to revenue. It’s unbelievable. [00:00:00] S0: It’s a many percentage points kind of thing? [00:00:00] S1: It’s extremely high. [00:00:00] S0: It’s many tens of percentage points for some of these private companies. So I am a little curious. Right now we’re in that part of the journey with AI coding where it’s just so valuable that people are like, “I have no choice but to spend, because it’s too productive.” I do think we can’t spend 70% of revenue on AI tokens forever and have valuable companies. So how do you think this might go? [00:00:00] S1: It’s complex. I’m very grateful for the stage Shopify’s at, because we are a profitable, public, trusted company. Right now, we really, really like the tokens we are buying. They’re incredibly valuable, and we are doing incredible things with them. It accelerates us in roadmap, and therefore in ambition. I think it just causes so many good things. [00:00:00] S0: You’re getting leverage on your spend. [00:00:00] S1: It’s a no-brainer. I have a very high opinion of markets. Markets are extremely good. They will figure out what the correct clearing price for these tokens is, and right now there’s few providers and maybe there’ll be more in the future. There’s all sorts of interesting moves around distillation and so on. I think companies will know how to wield these tools within their budgets. We are actually doing a good deal of this kind of thing as well, but we are still charging ahead because frankly, we like the tokens we are buying. It’s that simple.
· Tobi Lutke, Jack Altman
[00:00:00] S1: It’s a no-brainer. I have a very high opinion of markets. Markets are extremely good. They will figure out what the correct clearing price for these tokens is, and right now there’s few providers and maybe there’ll be more in the future. There’s all sorts of interesting moves around distillation and so on. I think companies will know how to wield these tools within their budgets. We are actually doing a good deal of this kind of thing as well, but we are still charging ahead because frankly, we like the tokens we are buying. It’s that simple. [00:00:00] S0: I agree, and my instinct is that it should go down. But there are possible worlds where price goes up. [00:00:00] S1: I agree. Honestly, no one knows. We are 10x-ing the amount of tokens we want every year right now, and I’m sure that’s going to go up. We are 3x-ing the amount of GPUs that we are putting into the world. Those lines are not converging anywhere good at price savings. [00:00:00] S0: Given that AI coding is dominating so much, what are the biggest changes to team design? There used to be this EPD triad, and it worked a certain way. There were certain ways that roadmaps were built and reviews happened, and people would go out and talk to the customers and bring that back. What is it now? What’s the most fundamental molecule?
· Tobi Lutke, Jack Altman
[00:00:00] S1: The small team is my bet. The 3-5 people team, which, funnily enough, has always been Shopify’s bias. This is quite gratifying. The reason why we had to very often go past it now is because you just need a lot of specialized skills, at least for certain moments, on the teams. Your example is a great one: talk to the customers. It’s having someone who actually does the customer research and talks with people. What we always did is, we co-located people from the support org. We routed them all the tickets and put them on the teams, which was an amazing way to do it. Now, the agentic harness around our teams is actually routing really good summarizations of what our customers say automatically back. So that’s now available to everyone, and then everyone can do more. Everyone is a 7 out of 10 on every skill now. That’s really helpful because it allows you to make teams smaller. The thing that we are working the most and thinking the most around at Shopify… I’m big on pace. Pace has to be induced, otherwise it’s received. Parkinson’s Law is one of my most recommended books, and I have a 1970s, or ‘60s, copy of it that I give to everyone, my executives. I own many of them of the original run, because it’s so meaningful. [00:00:00] S0: And the law is basically…? [00:00:00] S1: “Work expands to the time allocated to it.” The book is 60 pages and full of these kinds of wisdoms. This is the most important one. [00:00:00] S0: So one of the most important functions of a leader is to just compress time windows.